1. Introduction

Due to its low optical propagation loss and low-cost, flexible fabrication, silica glass is, without challenge, the number-one base material of photonic systems. Apart from passive transmission, important commercial applications of silica-based glasses - in fibres or in integrated devices -also include, e.g., amplification using rare-earth doping or the Raman effect, filtering with the use of UV-induced gratings, and switching through the thermo-optic effect. With the use of thermal poling, this line of applications is foreseen to be extended to include new functionality such as frequency mixing and linear electro-optic modulation/switching.

The poling-induced effective nonlinear susceptibility is normally assumed to be given by ${\mathrm{\chi }}_{\text{eff}}^{\left(2\right)}$ = 3E _built-inχ⁽³⁾, i.e., it comes from the intrinsic χ⁽³⁾ susceptibility of the glass and a permanent, built-in field distribution E _built-in that is created during the poling process due to displacement of charged species. In silica, peak values (local values) of ${\mathrm{\chi }}_{\text{eff}}^{\left(2\right)}$ up to ~ 1 pm/V are routinely reported [1, 2, 3], and recently ~ 3 pm/V was reported in Ge:SiO₂ thin films [4]. For exploitation in a device, however, one is concerned with the ${\mathrm{\chi }}_{\text{eff}}^{\left(2\right)}$ value integrated over the profile of the relevant guided mode or modes – e.g., the fundamental (F) and second-harmonic (SH) modes. Wavelength mixing, such as second-harmonic generation (SHG), is most favourably done with quasi-phase matching (QPM), and for this purpose it is, furthermore, imperative to optimise the confinement of the light and the spatial modulation of ${\mathrm{\chi }}_{\text{eff}}^{\left(2\right)}$ along the waveguide. Compared to nonlinear crystals, e.g., LiNbO₃ or KTP, glasses have the advantage of high damage threshold, low propagation loss, and small dispersion, making it attractive to perform QPM conversion of high-power or pulsed lasers in guided structures of centimeter lengths.

In this letter we show that planar silica-on-silicon technology can be used to realise periodically poled frequency-doublers. We employ silicon oxynitride (SiON) buried waveguides with soft top cladding consisting of borophosphosilicate glass (BPSG). As we will show, the planar platform allows one to obtain good control over the quasi phase-matching wavelength and bandwidth of the devices. A record-high normalised conversion efficiency is demonstrated.

2. Experimental

2.1. Fabrication

 

Fig. 1. Schematic of a chip with QPM devices. In this example, the chip contains three waveguides with different QPM periods.

Download Full Size | PDF

A chip with QPM devices is shown in Fig. 1. Of several glass combinations and waveguide geometries tested [5], the following recipe was found to be best suited for periodic poling: SiON core glass is deposited on a thermally oxidised, 4” Si substrate using plasma-enhanced chemical vapour deposition (PECVD) with reactant gases SiH₄, N₂O, and NH₃, and annealed at 1100 °C for 4 h in N₂ ambient. The index step at 633 nm is 0.010–0.025, depending on the recipe. Photolithographic masking and reactive ion-etching (RIE) with CF₄ and CHF₃ is used to make waveguide ridges with intentionally oblique sidewalls. Soft top cladding glass (BPSG), index matched to silica, is deposited with PECVD and reflowed at 1050 °C for 8 h in N₂.

Because the sidewalls of the ridges are oblique and the etch is rather shallow (~ 0.5 μm), a high-quality interface is obtained with good contact between the top cladding and the core glass, and the top cladding is free of fissure voids and folds that would normally arise in the top cladding if it were grown over ridges with steep, vertical sidewalls. Imperfect, internal interfaces can be detrimental for the poling ability and poling stability since they allow for undesired charge transport [5] – even if the structures provide loss-less, high-quality waveguiding.

Periodic electrodes of Al are fabricated on the top cladding, using e-gun evaporation and liftoff. We ascertain that the electrodes do not give rise to propagation loss. Due to the planar layout, each electrode may cover hundreds of waveguides and contain any desired range of periods and/or duty cycles. The electrodes are finally encapsulated with a few µm of undoped PECVD silica which receives no post-anneal. This capping layer serves three purposes [5]: (i) It hinders horizontal charge migration that may otherwise take place at the glass/air interface between the segments of the periodic electrode during the poling. (ii) It enables the periodic field to have a high contrast since it separates the electrode and core layer from the (zero-potential) top surface, while the electrode is placed relatively close to the core layer. (iii) It inhibits long-term degradation of the poling-induced nonlinearity that may otherwise occur at room temperature through exchange of charges with the atmosphere.

Notice that periodic erasure of the poling with UV light is not necessary, though it can easily be performed directly through the electrodes to further improve the nonlinear contrast.

2.2. Poling and optical characterisation

The chip is heated on a hot plate in air, with the Si substrate grounded and high voltage connected to the periodic electrode. The rise time is ~ 20 min, and the poling temperature is maintained for 15 min before the hotplate is rapidly cooled with a fan. At ≥ 50 °C the high voltage is removed. For SHG measurements, light from a CW tuneable Ti:sapphire laser is coupled into a single-mode fibre, led through a polarisation controller and into the QPM waveguide. On the exit side, the light is collected in a multi mode fibre (core diameter 62.5 μm, assumed collection efficiency 100 %) and either led into a power meter (for measurement of the F power, typically 1-100 mW) or through a free-space coupler containing a filter glass that blocks the F light, and into an optical spectrum analyser (for measurement of the SH power). For estimation of the F and SH powers in the waveguide, P_ω and P_2ω , wavelength-dependent system losses are taken into account. The propagation loss in the waveguides is ignored since it is found to be below 1 dB/cm at the investigated F and SH wavelengths and the maximum chip length is 1.5 cm.

3. Results and discussion

3.1. Device characterisation

For a device of length L and QPM period Λ, the SH power vs. wavelength is expected to follow

where μ and ε₀ are the permeability and permittivity of free space. The wave vector mismatch is Δḱ = k _2ω - 2k_ω - m2π/Λ = (4π/λ) (n _2ω - n_ω )- m2π/Λ where m = 1 is the order of the grating, and n_ω and n _2ω are the wavelength-dependent effective indices for the particualar waveguide. At the centre wavelength of the device, λ₀, we have Δḱ = 0. A _ovl is an effective modal area that is calculated from the overlap between the F and the SH modes as in refs. [6, 7], it is typically 2-2.25 times larger than the 1/e area of the F mode. ${\mathrm{\chi }}_{\text{expl}}^{\left(2\right)}$ is the exploited susceptibility, i.e., it is the part of ${\mathrm{\chi }}_{\text{eff}}^{\left(2\right)}$ that is spatially modulated (with period Λ) and overlapped with the F and SH modes. For an idealised poling with square-like nonlinear contrast {${\mathrm{\chi }}_{33}^{\left(2\right)}$,0,${\mathrm{\chi }}_{33}^{\left(2\right)}$,0,…} and 50 % duty cycle, one would have ${\mathrm{\chi }}_{\text{expl}}^{\left(2\right)}$ = ${\mathrm{\chi }}_{33}^{\left(2\right)}$.

 

Fig. 2. SH intensity as a function of pump wavelength for a 3-mm long device.

Download Full Size | PDF

In Fig. 2 we show typical measured and calculated SH spectra, from a poled QPM device with L = 3 mm, Λ = 23.5 μm, and centre wavelength λ₀ ≈ 831 nm. In the calculations, n_ω (λ) and n _2ω(λ) were obtained from effective-index simulations by using the core and cladding refractive indices that were measured at 633 nm in test films, and assuming that both scale with wavelength according to the Sellmeier equation for undoped silica. Excellent agreement is obtained between the measured and calculated spectra over more than 20-nm wavelength range and intensities spanning three orders of magnitude. Devices of lengths from 1 to 9 mm were fabricated and found to perform well. To match λ₀ to that of the measured spectra, a minor correction of at most ~ 0.07 % is typically made to n _2ω(λ). The width of the central peak and the position of the side lobes are fitted to the measured spectra by varying L in the calculation; typically the agreement falls within 3 % of the real device length even for the longest devices.

These results show that the QPM devices, fabricated with lengths up to 9 mm, have high uniformity along the waveguide and perform perfect phase matching over the full length of the device. The bandwidth near 830 nm is L × Δλ ≈ 0.69 cm × nm (FWHM). Notice that at 1550 nm (Λ ~ 50 μm), as much as L × Δλ ≈ 12 cm × nm is expected.

In Fig. 3 we show measured and calculated centre wavelengths λ₀ of a range of devices on the same chip, with values of electrode period Λ varying from 17.5 to 25 μm. The devices were separated by up to a few cm, and the index step at 633 nm was 0.010. The calculated points are found from effective-index simulations, as described above, but with no free paramters. The small difference between calculations and measured points is due to the fact that the material refractive indices of the BPSG and SiON layers are not known precisely at λ and λ/2 but only at 633 nm. Over the span of almost 150 nm, the measured points lie extremely close (within ±0.08 nm) to an ad-hoc 2nd order polynomial fit. In fact, the maximum device-to-device variation for different devices across the same wafer, having the same Λ, is around ±0.15 nm for wafers with high index step (0.025) and even less for wafers with more lightly doped core layers.

 

Fig. 3. Centre wavelength vs. electrode period for different devices on the same chip.

Download Full Size | PDF

 

Fig. 4. Normalised conversion efficiency η_N as a function of poling temperature for different poling voltages. The poling duration was 15 minutes. The devices in the ’#504’ series have slightly optimised design relative to those in the ’#404’ series (thinner top cladding and core layer, and thicker buffer layer). The dashed line is a guide to the eye.

Download Full Size | PDF

These findings show that the centre wavelength of the poled QPM devices is well controlled, with very small variation over a full wafer.

3.2. Optimum poling

The best figure of merit for the periodic poling is the normalised conversion efficiency, η_N, defined as the conversion efficiency per unit of pump power and squared length of the device: η_N = (P _2ω/P_ω /P_ωL ²). In Fig. 4 we plot values of tjn obtained after poling at different temperatures and with different poling voltages. The highest value is η_N = 1.4 × 10^-3 %/W/cm², obtained with 365 °C and +2500 V The corresponding nonlinearity is ${\mathrm{\chi }}_{\text{expl}}^{\left(2\right)}$ = 0.13 pm/V. To our knowledge, the value of η_N is almost an order of magnitude higher than what has previously been achieved with periodic glass poling [6]. With this value, one can efficiently frequency-double a pulsed laser with kW peak power by using a device with a length of a few cm.

In the literature, periodic waveguide poling has been reported only for slab waveguides up to now [8,9,7]. Due to poor confinement and less efficient periodic poling, only modest values up to η_N ~ 4 × 10^-7 %/W/cm² have been realised [7]. For fibres, Pruneri et al. [6] achieved ~ 2×10^-4 %/W/cm² by using a D-shaped fibre with a lithographically defined periodic electrode.

3.3. Stability

The measurements of conversion efficiency were in each case made within a day after the poling. A slight decay is observed over the course of days or weeks, but the decay is slowed down significantly with increasing thickness of the capping glass. Low-quality, unannealed PECVD silica of thickness ~ 1, 5, or 10 μm was used. We conclude that the decay is caused by exchange of charges with the air and migration of charges through the capping glass, eventually making the layer with the periodic electrode act as an equipotential layer in close proximity to the waveguide. This conclusion is supported by the observation that the decay is somewhat accelerated if the top surface of the device is covered with conducting paste and kept in direct ground contact. Since each of the waveguide constituents (SiO₂, BPSG, and SiON) possess good stability when poled, it is expected that good stability of the device will be obtained if it is sealed with abetter capping layer, e.g., annealed silica, silicon nitride or a polymer.

3.4. Poling mechanism

Figure 4 shows that the optimum poling temperature is ~ 365 °C (for both signs of poling voltage) which is higher than the ~250-280 °C commonly used for silica-based fibres and bulk glasses for comparable poling durations [2, 3]. We find the same optimum temperature when poling with a uniform electrode [10]. For negative voltage, 25 min poling time was also tested; only modest variation was observed. Thus, the charge carrier that is predominant in the waveguide poling appears to have low mobility. (If the optimum temperature were primarily governed by the dimensions of the glass, one would rather have expected that the planar waveguides, having a total glass thickness of only ~ 10-12 μm, are poled optimally at a lower temperature than are fibres and bulk glasses). It is plausible that H ions play a dominant role: H has lower mobility than Na [11, 12] that is normally considered to be responsible for poling of fused silica, and H is known to be incorporated from the PECVD reactant gases, whereas the Na content is very low in the PECVD glass. Alternatively, the high optimum poling temperature may be related to the charge-trapping property of oxynitrides [10, 13]. Notice that the high poling temperature can be advantageous since it may signal high intrinsic stability of the poled PECVD glass.

4. Conclusion

We performed periodic poling of planar glass waveguides on Si, obtaining a normalised SH conversion efficiency η_N =1.4 × 10^-3 %/W/cm² which corresponded to an exploited nonlinearity ${\mathrm{\chi }}_{\text{expl}}^{\left(2\right)}$ = 0.13 pm/V Perfect QPM is obtained, and good reproduction of centre wavelengths is demonstrated between approximately 740 and 860 nm. Devices with length up to 9 mm were fabricated. The waveguide design had been chosen to be particularly suited for poling; we used a SiO2 lower cladding, a SiON waveguide core made with a shallow etch and oblique side-walls, and a soft (BPSG) top cladding. The poling was performed with embedded, periodic Al electrodes that had been deposited on the top cladding and encapsulated under a SiO₂ capping layer. This kind of device may find use in efficient frequency-doubling of pulsed, high peak-power lasers. Further improvement of the conversion efficiency may include optimisation of the waveguide geometry, the core glass composition (e.g., use of Ge doping [4], heavy-metal doping [14], or nanoclusters [15, 16]). Chirped electrodes may also easily be introduced.